dsclaude — Claude Code & Claude Desktop launchers for alternative backends

中文文档

A collection of launchers and configurators that point Claude Code and Claude Desktop at third-party model backends (DeepSeek, MiMo, Qwen, GLM, Kimi, Ark, LongCat, MiniMax, TokenHub, SiliconFlow, etc.).

Tools

Tool What it does Platform Backend
dsclaude Claude Code CLI launcher macOS / Linux DeepSeek
dsclaude.ps1 Claude Code CLI launcher Windows DeepSeek
mmclaude Claude Code CLI launcher macOS / Linux Xiaomi MiMo
mmclaude.ps1 Claude Code CLI launcher Windows Xiaomi MiMo
qwclaude Claude Code CLI launcher macOS / Linux Alibaba Cloud Bailian (Qwen)
qwclaude.ps1 Claude Code CLI launcher Windows Alibaba Cloud Bailian (Qwen)
glmclaude Claude Code CLI launcher macOS / Linux ZhipuAI GLM
glmclaude.ps1 Claude Code CLI launcher Windows ZhipuAI GLM
kmclaude Claude Code CLI launcher macOS / Linux Moonshot AI Kimi
kmclaude.ps1 Claude Code CLI launcher Windows Moonshot AI Kimi
arkclaude Claude Code CLI launcher macOS / Linux Volcengine Ark (doubao/Kimi/DeepSeek/GLM…)
arkclaude.ps1 Claude Code CLI launcher Windows Volcengine Ark (doubao/Kimi/DeepSeek/GLM…)
lcclaude Claude Code CLI launcher macOS / Linux Meituan LongCat
lcclaude.ps1 Claude Code CLI launcher Windows Meituan LongCat
mxclaude Claude Code CLI launcher macOS / Linux MiniMax
mxclaude.ps1 Claude Code CLI launcher Windows MiniMax
hyclaude Claude Code CLI launcher macOS / Linux Tencent TokenHub (HY3/DeepSeek/Kimi/GLM…)
hyclaude.ps1 Claude Code CLI launcher Windows Tencent TokenHub (HY3/DeepSeek/Kimi/GLM…)
sfclaude Claude Code CLI launcher macOS / Linux SiliconFlow (DeepSeek/Kimi/GLM/Qwen/Yi…)
sfclaude.ps1 Claude Code CLI launcher Windows SiliconFlow (DeepSeek/Kimi/GLM/Qwen/Yi…)
xclaude Unified auto-detect launcher macOS / Linux Any (auto-detects from API keys)
dsclaude-desktop Claude Desktop GUI configurator macOS DeepSeek
dsclaude-desktop.ps1 Claude Desktop GUI configurator Windows DeepSeek
mmclaude-desktop Claude Desktop GUI configurator macOS Xiaomi MiMo
mmclaude-desktop.ps1 Claude Desktop GUI configurator Windows Xiaomi MiMo
qwclaude-desktop Claude Desktop GUI configurator macOS Alibaba Cloud Bailian (Qwen)
qwclaude-desktop.ps1 Claude Desktop GUI configurator Windows Alibaba Cloud Bailian (Qwen)
glmclaude-desktop Claude Desktop GUI configurator macOS ZhipuAI GLM
glmclaude-desktop.ps1 Claude Desktop GUI configurator Windows ZhipuAI GLM
kmclaude-desktop Claude Desktop GUI configurator macOS Moonshot AI Kimi
kmclaude-desktop.ps1 Claude Desktop GUI configurator Windows Moonshot AI Kimi
arkclaude-desktop Claude Desktop GUI configurator macOS Volcengine Ark
arkclaude-desktop.ps1 Claude Desktop GUI configurator Windows Volcengine Ark
lcclaude-desktop Claude Desktop GUI configurator macOS Meituan LongCat
lcclaude-desktop.ps1 Claude Desktop GUI configurator Windows Meituan LongCat
mxclaude-desktop Claude Desktop GUI configurator macOS MiniMax
mxclaude-desktop.ps1 Claude Desktop GUI configurator Windows MiniMax
hyclaude-desktop Claude Desktop GUI configurator macOS Tencent TokenHub
hyclaude-desktop.ps1 Claude Desktop GUI configurator Windows Tencent TokenHub
sfclaude-desktop Claude Desktop GUI configurator macOS SiliconFlow
sfclaude-desktop.ps1 Claude Desktop GUI configurator Windows SiliconFlow
skills/deepseek-vision Vision skill (zero deps) macOS / Linux DashScope Qwen
dsvision-mcp Vision MCP server macOS / Linux DashScope Qwen

How it works

xxclaude script-logic diagram

Script-logic overview: the installer, the xclaude auto-detect dispatcher, the provider CLI launchers, the Claude Desktop configurators, and the Codex bridge — all sharing lib/common.sh. The PNG has the draw.io XML embedded, so opening it in draw.io recovers the editable diagram (source · 中文版).

Quick start on macOS

# One-line install (recommended):
curl -fsSL https://raw.githubusercontent.com/Agents365-ai/dsclaude/main/install.sh | bash

# Or user install (no sudo):
curl -fsSL https://raw.githubusercontent.com/Agents365-ai/dsclaude/main/install.sh | bash -s -- --user

# Or manual:
git clone https://github.com/Agents365-ai/dsclaude.git
cd dsclaude
chmod +x xclaude
./xclaude

Quick start on Windows

# One-line install (PowerShell as admin recommended):
irm https://raw.githubusercontent.com/Agents365-ai/dsclaude/main/install.ps1 | iex

# Install with auto-PATH:
irm https://raw.githubusercontent.com/Agents365-ai/dsclaude/main/install.ps1 | iex -AddToPath

# Then launch:
pwsh -File xclaude.ps1

xclaude — Unified auto-detect launcher

Don't remember which script to run? xclaude auto-detects your API key and delegates to the right backend.

# Set any one of these in your ~/.zshrc:
export DEEPSEEK_API_KEY=sk-...     # → dsclaude
export GLM_API_KEY=...              # → glmclaude
export KIMI_API_KEY=sk-...          # → kmclaude
export ARK_API_KEY=...              # → arkclaude
# ... or any other supported key

xclaude                  # auto-detect and launch
xclaude fast             # ...with fast/flash tier
xclaude long effort max  # ...with 1M context + max effort
xclaude kimi             # forwarded: on Ark → kimi model, etc.

All positional args (model aliases, fast, long, effort, etc.) are forwarded to the underlying launcher. If multiple keys are set, the first found wins; set XCLAUDE_PREFER=dsclaude to force a specific provider.

dsclaude — Claude Code on DeepSeek

Follows the DeepSeek Anthropic API guide.

export DEEPSEEK_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # add to ~/.zshrc

dsclaude                 # default: deepseek-v4-pro (full reasoning)
dsclaude fast            # deepseek-v4-flash (cheaper / faster)
dsclaude long            # request 1M context window
dsclaude long fast       # 1M + flash

Sets the DeepSeek-recommended env vars (ANTHROPIC_BASE_URL, model mappings, CLAUDE_CODE_EFFORT_LEVEL=max), and exposes the alternate model in Claude Code's /model picker. Override context window via DSCLAUDE_MAX_TOKENS and effort via DSCLAUDE_EFFORT.

Both models natively support 1M-token context. The [1m] suffix (e.g. deepseek-v4-pro[1m]) is required in Claude Code — dsclaude sets it automatically.

Windows (PowerShell 7+): pwsh -File ./dsclaude.ps1 (same flags).

mmclaude — Claude Code on Xiaomi MiMo

export MIMO_API_KEY=sk-xxxxxxxxxxxxxxxxxx       # pay-as-you-go
# or
export MIMO_API_KEY=tp-xxxxxxxxxxxxxxxxxx       # Token Plan

mmclaude                  # start on mimo-v2.5-pro
mmclaude fast             # start on mimo-v2.5 (cheaper / faster flash tier)
mmclaude update           # git pull

Auto-detects base URL from the key prefix (sk-* → public, tp-* → Token Plan); override with MIMO_BASE_URL. Main/opus/sonnet slots run mimo-v2.5-pro while the haiku and subagent tiers run mimo-v2.5 (flash); mmclaude fast flips the main model to flash, and the other tier is exposed in the /model picker for mid-session switching. Unsets ANTHROPIC_API_KEY (per MiMo docs). Override the tiers with MIMO_MODEL / MIMO_FLASH_MODEL.

Windows (PowerShell 7+): pwsh -File ./mmclaude.ps1 (same flags).

qwclaude — Claude Code on Alibaba Cloud Bailian (Qwen)

# One key per plan — each comes from a different Bailian console entry:
export DASHSCOPE_API_KEY=sk-xxxxxxxxxxxxxxxxxx      # pay-as-you-go
export DASHSCOPE_CP_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # Coding Plan
export DASHSCOPE_TP_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # Token Plan team edition

# macOS / Linux — three model tiers
qwclaude                  # pay-as-you-go, qwen3.7-max (Beijing)     → DASHSCOPE_API_KEY
qwclaude plus             # qwen3.7-plus (balanced, ~1/6 cost of max)
qwclaude max              # explicit qwen3.7-max
qwclaude fast             # qwen3.6-flash as the main model
qwclaude intl             # pay-as-you-go on the Singapore endpoint
qwclaude coding           # Coding Plan (qwen3.7-plus recommended)   → DASHSCOPE_CP_API_KEY
qwclaude coding plus      # Coding Plan, qwen3.7-plus (default)
qwclaude coding fast      # Coding Plan, qwen3.6-plus
qwclaude token            # Token Plan (qwen3.7-max)                 → DASHSCOPE_TP_API_KEY
qwclaude token plus       # Token Plan, qwen3.7-plus
qwclaude update           # git pull

# Windows (PowerShell 7+)
pwsh -File ./qwclaude.ps1 coding

Picks the base URL, model lineup, and API-key variable per billing plan and model tier. Three model tiers:

Tier Pay-as-you-go & Token Plan Coding Plan
max (default) qwen3.7-max (flagship) N/A
plus qwen3.7-plus (balanced, ~1/6 cost) qwen3.7-plus (recommended)
flash (subagent/haiku) qwen3.6-flash qwen3.6-plus

Main/opus/sonnet slots run the selected tier (default qwen3.7-max) while the haiku + subagent tiers always use the flash model. fast flips the main model to flash; plus flips to qwen3.7-plus. The other tier is exposed in the /model picker for mid-session switching. Sets CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 and unsets ANTHROPIC_API_KEY to keep traffic on Bailian. Override via QWEN_PLAN / QWEN_REGION / QWEN_MODEL / QWEN_PLUS_MODEL / QWEN_FLASH_MODEL / QWEN_BASE_URL.

The Windows port (qwclaude.ps1) requires PowerShell 7+ (winget install Microsoft.PowerShell) — run it with pwsh -File.

glmclaude — Claude Code on ZhipuAI GLM

Follows ZhipuAI's Anthropic-compatible API.

export GLM_API_KEY=your_zhipu_api_key           # add to ~/.zshrc
# Get your key at: https://open.bigmodel.cn/usercenter/apikeys

glmclaude                  # default: glm-5.2 (full reasoning)
glmclaude fast             # glm-4.7 (cheaper / faster)
glmclaude update           # git pull

Sets the ZhipuAI-recommended env vars (ANTHROPIC_BASE_URL, model mappings), and exposes the alternate model in Claude Code's /model picker. Main/opus/sonnet slots run glm-5.2 while the haiku and subagent tiers run glm-4.7; glmclaude fast flips the main model to flash, and the other tier is exposed in the /model picker for mid-session switching.

By default uses the China endpoint (https://open.bigmodel.cn/api/anthropic). For the international endpoint, set GLM_BASE_URL:

export GLM_BASE_URL=https://api.z.ai/api/anthropic
glmclaude

Override the tiers with GLM_MODEL / GLM_FLASH_MODEL.

Windows (PowerShell 7+): pwsh -File ./glmclaude.ps1 (same flags).

kmclaude — Claude Code on Moonshot AI Kimi

export KIMI_API_KEY=sk-xxxxxxxxxxxxxxxxxx       # add to ~/.zshrc
# Get your key at: https://platform.moonshot.cn/console/api-keys

kmclaude                  # start on kimi-k3 (flagship)
kmclaude fast             # start on kimi-k2.5 (cheaper / faster tier)
kmclaude update           # git pull

Uses the official Kimi Anthropic-compatible endpoint (https://api.moonshot.cn/anthropic). Main/opus/sonnet slots run kimi-k3 while the haiku and subagent tiers run kimi-k2.5; kmclaude fast flips the main model to k2.5, and the other tier is exposed in the /model picker for mid-session switching. Unsets ANTHROPIC_API_KEY to avoid credential shadowing. Override the tiers with KIMI_MODEL / KIMI_FLASH_MODEL and the base URL with KIMI_BASE_URL.

Windows (PowerShell 7+): pwsh -File ./kmclaude.ps1 (same flags).

dsclaude-desktop — Claude Desktop GUI configurator

One-command configurator for Claude Desktop's built-in Third-Party Inference feature (Developer menu), pre-filled for DeepSeek.

Prerequisites

  1. Claude Desktop installed (claude.ai/download)
  2. Developer Mode enabled (Help → Troubleshooting → Enable Developer Mode, once)
  3. DEEPSEEK_API_KEY environment variable set

Usage

export DEEPSEEK_API_KEY=sk-xxxxxxxxxxxxxxxxxx
./dsclaude-desktop        # configure and restart
./dsclaude-desktop -h     # help

Generates an entry under ~/Library/Application Support/Claude-3p/configLibrary/, sets it as appliedId in _meta.json, then restarts the app. Existing GUI-added entries are preserved.

Switching modes

Claude Desktop's launch chooser handles Anthropic ↔ Gateway switching natively — no --revert flag. Click your profile → Disconnect (or sign out), then pick the other option at next launch.

Classic Chat (claude.ai-style) is unavailable in Gateway mode — it depends on Anthropic-hosted features not exposed via the inference API. Switch back to Anthropic mode to use it.

Windows

$env:DEEPSEEK_API_KEY = "sk-xxxxxxxxxxxxxxxxxx"
pwsh ./dsclaude-desktop.ps1

Writes to %APPDATA%\Claude-3p\configLibrary\. > Untested by the maintainer — please open an issue if anything misbehaves.

Prerequisites: Claude Desktop installed (Store or standard), DeepSeek API key. Unlike macOS, Developer Mode is auto-enabled by the script — no manual GUI toggle needed.

Config path: %LOCALAPPDATA%\Claude-3p\configLibrary\ (for Store/MSIX installs, the script also writes to the sandboxed package path as a fallback).

Tested on Windows 11 with Claude Desktop 1.7196 (Windows Store, arm64).

mmclaude-desktop — Claude Desktop on Xiaomi MiMo

Same configurator as dsclaude-desktop, pre-filled for Xiaomi MiMo. Reads MIMO_API_KEY and auto-detects the base URL from the key prefix (tp-* → Token Plan, else pay-as-you-go; override with MIMO_BASE_URL). Configures mimo-v2.5-pro + mimo-v2.5.

export MIMO_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # or tp-... for Token Plan
./mmclaude-desktop        # configure and restart (macOS)
./mmclaude-desktop -h     # help

# Windows (PowerShell)
$env:MIMO_API_KEY = "sk-xxxxxxxxxxxxxxxxxx"
pwsh ./mmclaude-desktop.ps1

qwclaude-desktop — Claude Desktop on Alibaba Cloud Bailian (Qwen)

Same configurator, with per-plan base URL, models, and key variable. Pay-as-you-go (DASHSCOPE_API_KEY) and Token Plan (DASHSCOPE_TP_API_KEY) default to qwen3.7-max + qwen3.6-flash; Coding Plan (DASHSCOPE_CP_API_KEY) configures qwen3.7-plus + qwen3.6-plus. Add plus to use qwen3.7-plus for any plan.

export DASHSCOPE_API_KEY=sk-xxxxxxxxxxxxxxxxxx       # pay-as-you-go
export DASHSCOPE_CP_API_KEY=sk-xxxxxxxxxxxxxxxxxx    # Coding Plan
export DASHSCOPE_TP_API_KEY=sk-xxxxxxxxxxxxxxxxxx    # Token Plan

./qwclaude-desktop            # pay-as-you-go (Beijing), qwen3.7-max, then restart
./qwclaude-desktop plus       # pay-as-you-go, qwen3.7-plus
./qwclaude-desktop intl       # pay-as-you-go, Singapore endpoint
./qwclaude-desktop coding     # Coding Plan (qwen3.7-plus)
./qwclaude-desktop token      # Token Plan, qwen3.7-max
./qwclaude-desktop token plus # Token Plan, qwen3.7-plus

# Windows (PowerShell)
pwsh ./qwclaude-desktop.ps1 -Plan coding
pwsh ./qwclaude-desktop.ps1 -Plan token -ModelTier plus

The .ps1 Windows ports mirror dsclaude-desktop.ps1 and are untested by the maintainer — please open an issue if anything misbehaves. Both configurators set unstableDisableModelVerification so Claude Desktop accepts the non-Anthropic model names, and (like dsclaude-desktop) disable Chat mode while the gateway is active.

glmclaude-desktop — Claude Desktop on ZhipuAI GLM

Same configurator, pre-filled for ZhipuAI GLM. Reads GLM_API_KEY. Configures glm-5.2 + glm-4.7.

export GLM_API_KEY=your_zhipu_api_key
./glmclaude-desktop        # configure and restart (macOS)
./glmclaude-desktop -h     # help

# Windows (PowerShell)
$env:GLM_API_KEY = "your_zhipu_api_key"
pwsh ./glmclaude-desktop.ps1

Override the base URL with GLM_BASE_URL (default: https://open.bigmodel.cn/api/anthropic).

kmclaude-desktop — Claude Desktop on Moonshot AI Kimi

Same configurator, pre-filled for Moonshot Kimi. Reads KIMI_API_KEY. Configures kimi-k3 + kimi-k2.5.

export KIMI_API_KEY=sk-xxxxxxxxxxxxxxxxxx
./kmclaude-desktop        # configure and restart (macOS)
./kmclaude-desktop -h     # help

# Windows (PowerShell)
$env:KIMI_API_KEY = "sk-xxxxxxxxxxxxxxxxxx"
pwsh ./kmclaude-desktop.ps1

Override the base URL with KIMI_BASE_URL (default: https://api.moonshot.cn/anthropic).

arkclaude — Claude Code on Volcengine Ark (Coding Plan)

Volcengine Ark's Coding Plan aggregates multiple model families under one subscription — doubao, Kimi, DeepSeek, GLM, MiniMax, and more. One API key, one endpoint, many models.

export ARK_API_KEY=your_ark_api_key           # add to ~/.zshrc
# Get your key at: https://console.volcengine.com/ark

# Three doubao tiers + provider aliases for quick model switching
arkclaude                  # doubao-seed-2.0-code (code specialist, default)
arkclaude plus             # doubao-seed-2.0-pro (balanced flagship)
arkclaude fast             # doubao-seed-2.0-lite (cheap / fast)
arkclaude kimi             # kimi-k2.7-code (Kimi code model)
arkclaude kimi-pro         # kimi-k2.6
arkclaude deepseek         # deepseek-v4-pro
arkclaude deepseek-flash   # deepseek-v4-flash
arkclaude glm              # glm-5.2
arkclaude minimax          # minimax-m2.7
arkclaude update           # git pull

Uses the Ark Coding Plan endpoint (https://ark.cn-beijing.volces.com/api/coding). Main/opus/sonnet slots run the chosen model while the haiku + subagent tiers always use the cheapest model in the same family (or doubao-seed-2.0-lite as fallback). The /model picker exposes a smart alternate based on your selection — Pro ↔ Code when using doubao, K2.6 ↔ K2.7-code when using Kimi, etc. Sets CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 and unsets ANTHROPIC_API_KEY to keep traffic on Ark. Override via ARK_MODEL / ARK_PLUS_MODEL / ARK_FLASH_MODEL / ARK_BASE_URL.

Windows (PowerShell 7+): pwsh -File ./arkclaude.ps1 (same flags).

arkclaude-desktop — Claude Desktop on Volcengine Ark

Same configurator, pre-filled for Ark Coding Plan. Reads ARK_API_KEY. Supports all model aliases for picking the inference model.

export ARK_API_KEY=your_ark_api_key
./arkclaude-desktop         # configure with doubao-seed-2.0-code and restart
./arkclaude-desktop plus    # doubao-seed-2.0-pro
./arkclaude-desktop kimi    # kimi-k2.7-code
./arkclaude-desktop deepseek # deepseek-v4-pro
./arkclaude-desktop glm     # glm-5.2
./arkclaude-desktop minimax # minimax-m2.7
./arkclaude-desktop -h      # help

# Windows (PowerShell)
$env:ARK_API_KEY = "your_ark_api_key"
pwsh ./arkclaude-desktop.ps1 -ModelTier kimi

Override the base URL with ARK_BASE_URL (default: https://ark.cn-beijing.volces.com/api/coding).

lcclaude — Claude Code on Meituan LongCat

LongCat-2.0 is a 1.6T-param MoE model built entirely on Chinese-manufactured compute. It ranked #2 globally in Claude Code monthly call volume on OpenRouter. Free during public beta — 500K tokens/day.

export LONGCAT_API_KEY=lc-xxxxxxxxxxxxxxxxxx   # add to ~/.zshrc
# Get your free key at: https://longcat.chat/platform

lcclaude                  # LongCat-2.0 (1M context, flagship)
lcclaude fast             # LongCat-Flash-Chat (cheaper / faster)
lcclaude think            # LongCat-Flash-Thinking (deep reasoning)
lcclaude update           # git pull

Uses the LongCat Anthropic-compatible endpoint (https://api.longcat.chat/anthropic). Main/opus/sonnet slots run the chosen model while the haiku + subagent tiers use LongCat-Flash-Chat. The /model picker exposes the alternate tier. Override via LONGCAT_MODEL / LONGCAT_FLASH_MODEL / LONGCAT_THINK_MODEL / LONGCAT_BASE_URL.

Windows (PowerShell 7+): pwsh -File ./lcclaude.ps1 (same flags).

lcclaude-desktop — Claude Desktop on Meituan LongCat

Same configurator, pre-filled for LongCat. Reads LONGCAT_API_KEY. Configures LongCat-2.0 + LongCat-Flash-Chat.

export LONGCAT_API_KEY=lc-xxxxxxxxxxxxxxxxxx
./lcclaude-desktop        # configure and restart (macOS)
./lcclaude-desktop -h     # help

# Windows (PowerShell)
$env:LONGCAT_API_KEY = "lc-xxxxxxxxxxxxxxxxxx"
pwsh ./lcclaude-desktop.ps1

Override the base URL with LONGCAT_BASE_URL (default: https://api.longcat.chat/anthropic).

mxclaude — Claude Code on MiniMax

MiniMax-M3 is a 1M-context flagship model optimized for agentic reasoning, tool use, and coding. Requires a Coding Plan (Token Plan) API key — not pay-as-you-go.

export MINIMAX_API_KEY=your_minimax_api_key    # add to ~/.zshrc
# Get your key at: https://platform.minimaxi.com

mxclaude                  # MiniMax-M3 (1M context, flagship)
mxclaude fast             # MiniMax-M2.5 (cheaper / faster)
mxclaude long             # request 1M context window
mxclaude effort max       # set reasoning effort (low|medium|high|xhigh|max)
mxclaude update           # git pull

Uses the MiniMax Anthropic-compatible endpoint (default: https://api.minimaxi.com/anthropic for China; override with MINIMAX_BASE_URL for international https://api.minimax.io/anthropic). Main/opus/sonnet slots run the chosen model while the haiku + subagent tiers use MiniMax-M2.5. The /model picker exposes the alternate tier. Override via MINIMAX_MODEL / MINIMAX_FLASH_MODEL / MINIMAX_BASE_URL / MINIMAX_CTX / MINIMAX_OUTPUT / MINIMAX_EFFORT.

Windows (PowerShell 7+): pwsh -File ./mxclaude.ps1 (same flags).

mxclaude-desktop — Claude Desktop on MiniMax

Same configurator, pre-filled for MiniMax. Reads MINIMAX_API_KEY. Configures MiniMax-M3 + MiniMax-M2.5.

export MINIMAX_API_KEY=your_minimax_api_key
./mxclaude-desktop        # configure and restart (macOS)
./mxclaude-desktop -h     # help

# Windows (PowerShell)
$env:MINIMAX_API_KEY = "your_minimax_api_key"
pwsh ./mxclaude-desktop.ps1

Override the base URL with MINIMAX_BASE_URL (default: https://api.minimaxi.com/anthropic; international: https://api.minimax.io/anthropic).

hyclaude — Claude Code on Tencent TokenHub

Tencent TokenHub is a multi-model aggregator — one subscription gives access to Hunyuan (HY3), DeepSeek, GLM, Kimi, MiniMax, Qwen, and more.

export HY_API_KEY=your_api_key             # add to ~/.zshrc
# Get your key at: https://cloud.tencent.com/product/lkeap

hyclaude                  # hy3-preview (Tencent default)
hyclaude fast             # hy-mt2-lite (cheap / fast)
hyclaude deepseek         # deepseek-v4-pro
hyclaude kimi             # kimi-k2.7-code
hyclaude glm              # glm-5.2
hyclaude minimax          # minimax-m3
hyclaude qwen             # qwen3.5-plus
hyclaude long             # request max context window
hyclaude effort max       # set reasoning effort (low|medium|high|xhigh|max)
hyclaude update           # git pull

Uses the TokenHub Anthropic-compatible endpoint (default: https://api.lkeap.cloud.tencent.com/plan/anthropic). Main/opus/sonnet slots run the chosen model while haiku + subagent tiers use hy-mt2-lite (or a same-family flash model for non-HY selections). The /model picker exposes a smart alternate. Override via HY_MODEL / HY_FLASH_MODEL / HY_BASE_URL / HY_CTX / HY_OUTPUT / HY_EFFORT.

Windows (PowerShell 7+): pwsh -File ./hyclaude.ps1 (same flags).

hyclaude-desktop — Claude Desktop on Tencent TokenHub

Same configurator, pre-filled for TokenHub. Reads HY_API_KEY. Supports model aliases for the backend model.

export HY_API_KEY=your_api_key
./hyclaude-desktop         # configure with hy3-preview and restart
./hyclaude-desktop kimi    # kimi-k2.7-code
./hyclaude-desktop deepseek # deepseek-v4-pro
./hyclaude-desktop glm     # glm-5.2
./hyclaude-desktop minimax # minimax-m3
./hyclaude-desktop qwen    # qwen3.5-plus
./hyclaude-desktop -h      # help

# Windows (PowerShell)
$env:HY_API_KEY = "your_api_key"
pwsh ./hyclaude-desktop.ps1 -ModelTier kimi

Override the base URL with HY_BASE_URL (default: https://api.lkeap.cloud.tencent.com/plan/anthropic).

sfclaude — Claude Code on SiliconFlow

SiliconFlow (硅基流动) is a multi-model platform aggregating DeepSeek, Kimi, GLM, MiniMax, Qwen, Yi, and more. Models use the provider/Model-Name convention (e.g. deepseek-ai/DeepSeek-V4-PRO).

export SF_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # add to ~/.zshrc
# Get your key at: https://cloud.siliconflow.cn/account/ak

sfclaude                  # deepseek-ai/DeepSeek-V4-PRO (default)
sfclaude fast             # deepseek-ai/DeepSeek-V3 (cheaper)
sfclaude kimi             # moonshotai/Kimi-K2-Instruct-0905
sfclaude glm              # Pro/zai-org/GLM-5
sfclaude minimax          # Pro/MiniMaxAI/MiniMax-M2.5
sfclaude qwen             # Qwen/Qwen2.5-Coder
sfclaude yi               # 01-ai/Yi-1.5
sfclaude r1               # deepseek-ai/DeepSeek-R1 (deep reasoning)
sfclaude long             # request max context window
sfclaude effort max       # set reasoning effort (low|medium|high|xhigh|max)
sfclaude update           # git pull

Uses the SiliconFlow Anthropic-compatible endpoint (https://api.siliconflow.cn/). Main/opus/sonnet slots run the chosen model while haiku + subagent tiers use deepseek-ai/DeepSeek-V3. Override via SF_MODEL / SF_FLASH_MODEL / SF_BASE_URL / SF_CTX / SF_OUTPUT / SF_EFFORT.

Windows (PowerShell 7+): pwsh -File ./sfclaude.ps1 (same flags).

sfclaude-desktop — Claude Desktop on SiliconFlow

Same configurator, pre-filled for SiliconFlow. Reads SF_API_KEY. Supports model aliases.

export SF_API_KEY=sk-xxxxxxxxxxxxxxxxxx
./sfclaude-desktop         # deepseek-ai/DeepSeek-V4-PRO and restart
./sfclaude-desktop kimi    # moonshotai/Kimi-K2-Instruct-0905
./sfclaude-desktop r1      # deepseek-ai/DeepSeek-R1
./sfclaude-desktop -h      # help

# Windows (PowerShell)
$env:SF_API_KEY = "sk-xxxxxxxxxxxxxxxxxx"
pwsh ./sfclaude-desktop.ps1 -ModelTier kimi

deepseek-vision skill — Vision (zero-dependency)

Gives text-only agents (like DeepSeek) the ability to "see" images. When the agent encounters an image, it calls analyze-image, which sends it to Qwen3.6-Flash and returns a text description.

export DASHSCOPE_API_KEY=sk-xxxxxxxxxxxxxxxxxx
./skills/deepseek-vision/analyze-image /path/to/screenshot.png "What error is shown?"
./skills/deepseek-vision/analyze-image https://example.com/diagram.png

Works with any agent that loads SKILL.md (Claude Code, Cowork, etc.). Default model qwen3.6-flash; override via DSVISION_MODEL and DSVISION_BASE_URL.

Limitation: requires a file path or URL — inline images (drag-drop, paste, "+ → Add files or photos") aren't supported. Use dsvision-mcp below for that.

dsvision-mcp — Vision (MCP server)

Same functionality as the skill above, but runs as an MCP server — bypassing two Cowork sandbox limitations:

  1. Network egress — the skill's DashScope API calls are firewalled inside Cowork's VM; the MCP server runs outside it
  2. Inline images — auto-picks the latest cached image from ~/.claude/image-cache/, so drag-drop/paste/"+" workflows work (macOS only; Windows Cowork doesn't cache inline images to disk)

Setup

pip3 install fastmcp requests
export DASHSCOPE_API_KEY=sk-xxxxxxxxxxxxxxxxxx   # add to ~/.zshrc
cd /path/to/dsclaude && pwd    # note the absolute path

Add to the MCP config file matching your mode:

Mode Config file
3P/Gateway (DeepSeek via dsclaude-desktop) ~/Library/Application Support/Claude-3p/claude_desktop_config.json
Standard Anthropic ~/Library/Application Support/Claude/claude_desktop_config.json 
{
  "mcpServers": {
    "dsvision": {
      "command": "/absolute/path/dsclaude/dsvision-mcp"
    }
  }
}

Restart Claude Desktop. The analyze_image tool appears automatically.

Usage

analyze_image()                           # auto: latest cached image
analyze_image(image_path="/abs/path/foo.png")
analyze_image(focus="What error is shown?")

Troubleshooting

Symptom Check
Tool doesn't appear Wrong config file path / invalid JSON (validate with python3 -m json.tool)
Tool errors DASHSCOPE_API_KEY not set
ModuleNotFoundError Use pip3 not pip
Image not found Pass absolute path, or check ~/.claude/image-cache/ exists

Skill vs MCP: which to use

Scenario Use
Claude Code (CLI), explicit paths skills/deepseek-vision (zero deps)
Cowork / Desktop with inline images dsvision-mcp (only option that works)
Cowork with explicit paths, sandbox tweaks OK either

Support

If these scripts save you time, consider supporting the author:

Author

Agents365-ai · Bilibili · GitHub

License

CC BY-NC 4.0 — free for non-commercial use. Commercial use requires permission.